rsr.rsr.acquisition_score¶

rsr.rsr.acquisition_score(d_upper, d_lower, gamma=1.0)[source]¶

Acquisition score (d+ + d-) - gamma * |d+ - d-|.

gamma = 0 gives pure exploration, i.e. the candidate furthest from both frontiers wins. Larger gamma increasingly favours balanced candidates, whose two deficits are close and whose outcome is therefore least predictable.

Parameters:

d_upper (Tensor) – Upper deficits.
d_lower (Tensor) – Lower deficits.
gamma (float) – Non-negative weight balancing exploration against uncertainty.

Return type:

Tensor

Returns:

Float64 tensor of scores, same shape as the deficits.